fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-29 22:30:24 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

freedreno/a6xx, turnip: Set CONSTANTRAMMODE correctly

Browse source 
This should fix hangs when using more than 256 constants on a7xx.

Fixes: 5879eaac18 ("ir3: Increase compute const size on a7xx")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34746>
(cherry picked from commit 80bcbc0e92)
This commit is contained in:
Connor Abbott 2025-04-28 19:00:07 -04:00 • committed by Eric Engestrom
parent 2a06a20a4a
commit 385a56642b
4 changed files with 19 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -64,7 +64,7 @@
"description": "freedreno/a6xx, turnip: Set CONSTANTRAMMODE correctly",
"nominated": true,
"nomination_type": 2,
"resolution": 0,
"resolution": 1,
"main_sha": null,
"because_sha": "5879eaac185ed1c167fd01aff9b91c7cbe43ab0a",
"notes": null

8
src/freedreno/computerator/a6xx.cc
View file

@ -180,14 +180,18 @@ cs_program_emit(struct fd_ringbuffer *ring, struct kernel *kernel)
}
uint32_t shared_size = MAX2(((int)v->shared_size - 1) / 1024, 1);
enum a6xx_const_ram_mode mode =
v->constlen > 256 ? CONSTLEN_512 :
(v->constlen > 192 ? CONSTLEN_256 :
(v->constlen > 128 ? CONSTLEN_192 : CONSTLEN_128));
OUT_PKT4(ring, REG_A6XX_SP_CS_CTRL_REG1, 1);
OUT_RING(ring, A6XX_SP_CS_CTRL_REG1_SHARED_SIZE(shared_size) |
A6XX_SP_CS_CTRL_REG1_CONSTANTRAMMODE(CONSTLEN_256));
A6XX_SP_CS_CTRL_REG1_CONSTANTRAMMODE(mode));
if (CHIP == A6XX && a6xx_backend->info->a6xx.has_lpac) {
OUT_PKT4(ring, REG_A6XX_HLSQ_CS_CTRL_REG1, 1);
OUT_RING(ring, A6XX_HLSQ_CS_CTRL_REG1_SHARED_SIZE(1) |
A6XX_HLSQ_CS_CTRL_REG1_CONSTANTRAMMODE(CONSTLEN_256));
A6XX_HLSQ_CS_CTRL_REG1_CONSTANTRAMMODE(mode));
}
uint32_t local_invocation_id, work_group_id;

8
src/freedreno/vulkan/tu_shader.cc
View file

@ -1530,14 +1530,18 @@ tu6_emit_cs_config(struct tu_cs *cs,
tu6_emit_xs(cs, MESA_SHADER_COMPUTE, v, pvtmem, binary_iova);
uint32_t shared_size = MAX2(((int)v->shared_size - 1) / 1024, 1);
enum a6xx_const_ram_mode mode =
v->constlen > 256 ? CONSTLEN_512 :
(v->constlen > 192 ? CONSTLEN_256 :
(v->constlen > 128 ? CONSTLEN_192 : CONSTLEN_128));
tu_cs_emit_pkt4(cs, REG_A6XX_SP_CS_CTRL_REG1, 1);
tu_cs_emit(cs, A6XX_SP_CS_CTRL_REG1_SHARED_SIZE(shared_size) |
A6XX_SP_CS_CTRL_REG1_CONSTANTRAMMODE(CONSTLEN_256));
A6XX_SP_CS_CTRL_REG1_CONSTANTRAMMODE(mode));
if (CHIP == A6XX && cs->device->physical_device->info->a6xx.has_lpac) {
tu_cs_emit_pkt4(cs, REG_A6XX_HLSQ_CS_CTRL_REG1, 1);
tu_cs_emit(cs, A6XX_HLSQ_CS_CTRL_REG1_SHARED_SIZE(shared_size) |
A6XX_HLSQ_CS_CTRL_REG1_CONSTANTRAMMODE(CONSTLEN_256));
A6XX_HLSQ_CS_CTRL_REG1_CONSTANTRAMMODE(mode));
}
uint32_t local_invocation_id =

8
src/gallium/drivers/freedreno/a6xx/fd6_compute.cc
View file

@ -220,14 +220,18 @@ fd6_launch_grid(struct fd_context *ctx, const struct pipe_grid_info *info) in_dt
uint32_t shared_size =
MAX2(((int)(cs->v->cs.req_local_mem + info->variable_shared_mem) - 1) / 1024, 1);
enum a6xx_const_ram_mode mode =
cs->v->constlen > 256 ? CONSTLEN_512 :
(cs->v->constlen > 192 ? CONSTLEN_256 :
(cs->v->constlen > 128 ? CONSTLEN_192 : CONSTLEN_128));
OUT_PKT4(ring, REG_A6XX_SP_CS_CTRL_REG1, 1);
OUT_RING(ring, A6XX_SP_CS_CTRL_REG1_SHARED_SIZE(shared_size) |
A6XX_SP_CS_CTRL_REG1_CONSTANTRAMMODE(CONSTLEN_256));
A6XX_SP_CS_CTRL_REG1_CONSTANTRAMMODE(mode));
if (CHIP == A6XX && ctx->screen->info->a6xx.has_lpac) {
OUT_PKT4(ring, REG_A6XX_HLSQ_CS_CTRL_REG1, 1);
OUT_RING(ring, A6XX_HLSQ_CS_CTRL_REG1_SHARED_SIZE(shared_size) |
A6XX_HLSQ_CS_CTRL_REG1_CONSTANTRAMMODE(CONSTLEN_256));
A6XX_HLSQ_CS_CTRL_REG1_CONSTANTRAMMODE(mode));
}
const unsigned *local_size =